Anthropic Opens Beta of "Reflect," a Feature for Reviewing Your Claude Usage
On July 9, Anthropic released a new feature called "Reflect" in beta, designed to help users look back on and visualize how they use Claude. Available from Settings, the feature reviews your chat activity with Claude over the past 1 to 12 months and provides a dashboard summarizing key topics, usage patterns, and the kinds of tasks Claude handles well.
What sets it apart is that it goes beyond simply displaying statistics: it is designed to prompt users to ask themselves "how do I want to bring AI into my life?" It periodically surfaces questions like "What do I want to keep doing with my own hands?" and users can discuss them with Claude if they choose. It also includes tools for self-managing usage, such as quiet hours settings and notifications that encourage breaks after a certain amount of time.
On the privacy front, secret chats and source files from connected tools are excluded, and conversations related to medical integrations are completely left out of the insights. The feature was reportedly developed in collaboration with experts including the MIT Media Lab and Digital Wellness Lab.
TechCrunch, however, notes that by making users' dependence on Claude visible, the feature may actually work to further entrench AI in everyday life. The argument goes that showing users how much they use AI reminds them of its value and ultimately drives engagement higher.
For now, the feature is available to Free, Pro, and Max users who have the Memory feature turned on.
OpenAI Finds About 30% of SWE-Bench Pro Is Broken, Withdraws Its Recommendation
OpenAI reviewed SWE-Bench Pro, a widely used benchmark for measuring the programming ability of AI models, and found that roughly 30% of its tasks are flawed. The company has withdrawn its earlier recommendation of the benchmark.
The episode highlights just how difficult it is to evaluate AI coding ability. When a benchmark's task design itself is flawed, model scores may not accurately reflect real-world capability. The fact that OpenAI discovered the problem itself signals that major providers are starting to take benchmark reliability seriously.
Other coverage on July 9 had already raised "doubts about the reliability of coding benchmarks," suggesting that benchmark issues are increasingly being recognized as an industry-wide problem.
Anthropic, OpenAI, and SpaceX IPOs Would Exceed 25 Years of Total Tech Exits
According to a TechCrunch analysis, IPOs by Anthropic, OpenAI, and SpaceX are on track to surpass the combined value of all U.S. VC-backed exits (IPOs, acquisitions, and the like) since 2000.
The figure illustrates just how enormous the three companies' valuations are, showing that capital concentration in AI has reached an unprecedented level. Anthropic and OpenAI as AI companies, and SpaceX as a space and transportation company, operate in different fields — yet each stands as an IPO candidate on a scale that would dwarf the tech industry's entire exit value.
Whether this situation is sustainable depends on future market conditions, but at least for now, the scale of the AI bubble has been borne out statistically.
Meta's First Paid AI "Muse Spark 1.1" as Zuckerberg Pledges "Aggressive Pricing"
According to Bloomberg, Meta has released its first paid AI model, "Muse Spark 1.1," with CEO Mark Zuckerberg declaring that the company will pursue "aggressive pricing." Described as an agentic model, Muse Spark 1.1 is Meta's first AI product that users will pay to use.
Meta's AI features have been free until now, but intensifying competition and monetization pressure appear to have pushed the company toward paid tiers. The phrase "aggressive pricing" may signal a strategy of undercutting competitors to capture market share.
LinkedIn and X Overrun by AI Spam, Browsing Data Confirms
According to 404media, spam from AI-generated content has surged on LinkedIn and X (formerly Twitter), and browsing data backs up the reality of the problem.
The influx of machine-generated content is a major factor degrading the quality of information on social media platforms. On LinkedIn in particular, AI-generated posts disguised as professional content are on the rise, fueling growing concern that user trust is being eroded.
AI Agent Company Raises $100 Million Using AI Agents
Bloomberg reports that a startup that builds AI agents used its own AI agents to run a $100 million fundraising process. By putting its product — the AI agents themselves — to work in a real fundraising effort, the company effectively "demonstrated the effectiveness of its own technology."
It is a fascinating example of AI agents evolving from mere developer tools into practical tools that automate core business processes.
---